A method and apparatus for monitoring maladaptive thinking based on large language models

Abstract

The application provides a bad thinking monitoring method and device based on a large language model, wherein the method is applied to a conference system and comprises the following steps: monitoring group discussion, obtaining background information of a person participating in the group discussion and a conversation sequence generated by the group discussion; inputting the conversation sequence and the background information into a bad thinking monitoring model to obtain a thinking monitoring result output by the bad thinking monitoring model; the bad thinking monitoring model is obtained by training sample background information of a sample person participating in group discussion, a sample conversation sequence generated by sample group discussion and a bad thinking label of the sample group discussion on the basis of a large language model, and the method overcomes the defects of traditional schemes that cannot monitor group discussion and detect bad thinking in the group discussion, realizes monitoring of bad thinking in a multi-person conference, provides reference and basis for solving bad thinking of a group, can improve the comprehensiveness and accuracy of decision-making, and avoids the risk of decision-making failure.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and device for monitoring maladaptive thinking based on a large language model. Background Technology

[0002] During discussions, due to factors such as participants' lack of ability to propose high-quality solutions, differing interests within and outside the meeting, and time constraints, decisions are often made by the highest decision-maker or simply by the majority rule. This reveals undesirable thinking patterns such as group dependency, rigidity, and divergent thinking. Furthermore, solutions proposed in this manner are often based on limited information, thus carrying a significant risk of decision failure.

[0003] With the continuous development of artificial intelligence technology, natural language processing (NLP) has made significant progress in text processing. However, traditional NLP techniques mainly focus on text summarization, topic extraction, and machine translation, and their ability to handle multi-person, multi-turn dialogues in meetings is relatively limited. Furthermore, while traditional NLP techniques can assist in understanding and analyzing meeting content to some extent, they cannot solve the fundamental problem in multi-person meetings: the inability to effectively monitor and identify flawed thinking patterns that arise during decision-making. Summary of the Invention

[0004] This invention provides a method and apparatus for monitoring undesirable thinking based on a large-scale language model, which solves the shortcomings of existing technologies that cannot monitor group discussions as a whole and cannot detect undesirable thinking in group discussions, thereby enabling the monitoring of undesirable thinking in multi-person meetings.

[0005] This invention provides a method for detecting undesirable thought processes based on a large-scale language model, applied to a conference system. The method includes:

[0006] The group discussion is monitored, and background information of the participants in the group discussion and the conversation sequence generated by the group discussion are obtained. The conversation sequence includes the speeches, actions and corresponding times of the participants in the group discussion.

[0007] The conversation sequence and the background information are input into the negative thought monitoring model to obtain the thought monitoring results output by the negative thought monitoring model.

[0008] The negative thought monitoring model is trained based on a large language model, using the sample background information of participants in the sample group discussion, the sample conversation sequence generated by the sample group discussion, and the negative thought labels of the sample group discussion.

[0009] According to the present invention, a method for monitoring maladaptive thinking based on a large language model is provided, wherein the maladaptive thinking monitoring model includes a large language model and a target language model with identical structures, and the target language model is constructed based on the large language model;

[0010] The step of inputting the conversation sequence and the background information into the malfunctioning thought monitoring model to obtain the thought monitoring results output by the malfunctioning thought monitoring model includes:

[0011] The background information is input into the target language model in the negative thought monitoring model to obtain the background features output by the target language model.

[0012] The background features and the conversation sequence are input into the large language model in the malthought monitoring model to obtain the thought monitoring results output by the large language model.

[0013] According to the present invention, a method for detecting malfunctioning thoughts based on a large language model is provided, wherein the malfunctioning thought detection model further includes a feature fusion layer; the feature fusion layer connects the target language model and the large language model;

[0014] The process of inputting the background features and the conversation sequence into the large language model of the maladaptive thinking monitoring model to obtain the thinking monitoring results output by the large language model includes:

[0015] The background features and the conversation sequence are input into the feature fusion layer to obtain the fused features output by the feature fusion layer;

[0016] The fusion features are input into the large language model to obtain the thought monitoring results output by the large language model.

[0017] According to the present invention, a method for detecting malfunctioning thoughts based on a large language model is provided, wherein the malfunctioning thought detection model is trained based on the following steps:

[0018] Based on the large language model, a twin model is constructed, which includes a first language model and a second language model with the same parameters and structure.

[0019] The sample background information is input into the first language model to obtain the sample background features output by the first language model. The sample background features and the sample conversation sequence are input into the second language model to obtain the predictive thinking monitoring results output by the second language model.

[0020] Based on the predicted thinking monitoring results and the negative thinking labels, the first language model in the twin model is trained to obtain the negative thinking monitoring model.

[0021] According to the present invention, a method for monitoring maladaptive thinking based on a large language model is provided, wherein the first language model includes multiple first feature conversion layers and the second language model includes multiple second feature conversion layers;

[0022] The twin model also includes multiple initial feature fusion layers. The first language model and the second language model are connected through multiple initial feature fusion layers. The first feature transformation layer, the second feature transformation layer and the initial feature fusion layer have the same number of layers.

[0023] Any initial feature fusion layer connects the first feature transformation layer of the corresponding layer in the first language model to the second feature transformation layer of the layer above the corresponding layer in the second language model;

[0024] The initial feature fusion layer is used to fuse the output features of the first feature transformation layer of the corresponding layer in the first language model with the output features of the second feature transformation layer of the previous layer in the second language model, and use the fused features as the input features of the second feature transformation layer of the corresponding layer in the second language model.

[0025] According to the present invention, a method for monitoring malfunctioning thoughts based on a large language model is provided, wherein the first language model in the twin model is trained based on the predicted thought monitoring results and the malfunctioning thought labels to obtain a malfunctioning thought monitoring model, comprising:

[0026] Based on the predicted thinking monitoring results and the negative thinking labels, the first language model and each initial feature fusion layer in the twin model are trained to obtain the target language model and multiple feature fusion layers.

[0027] Based on the large-scale language model, the target language model, and the various feature fusion layers, a model for monitoring negative thinking is constructed.

[0028] According to a method for monitoring malfunctioning thought patterns based on a large language model provided by the present invention, the method further includes inputting the conversation sequence and the background information into the malfunctioning thought monitoring model to obtain the thought monitoring results output by the malfunctioning thought monitoring model.

[0029] When the thought monitoring results indicate the presence of negative thoughts in the group discussion, an alarm prompt and control tool for negative thoughts are obtained so that the participants in the group discussion can control the group discussion based on the alarm prompt and control tool for negative thoughts;

[0030] The negative thought warning and control tool is generated by the negative thought monitoring model based on the thought monitoring results.

[0031] This invention also provides a device for monitoring unhealthy thinking based on a large-scale language model, applied to a conference system, the device comprising:

[0032] The acquisition unit is used to monitor the group discussion and acquire the background information of the participants in the group discussion, as well as the conversation sequence generated by the group discussion. The conversation sequence includes the speeches and operations of the participants in the group discussion and their corresponding times.

[0033] A monitoring unit is used to input the conversation sequence and the background information into the negative thinking monitoring model to obtain the thinking monitoring results output by the negative thinking monitoring model.

[0034] The negative thought monitoring model is trained based on a large language model, using the sample background information of participants in the sample group discussion, the sample conversation sequence generated by the sample group discussion, and the negative thought labels of the sample group discussion.

[0035] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method for monitoring maladaptive thinking based on a large language model as described above.

[0036] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for monitoring maladaptive thinking based on a large language model as described above.

[0037] This invention provides a method and apparatus for monitoring undesirable thinking based on a large-scale language model. It monitors group discussions, acquires background information of participants, and obtains conversation sequences generated during the discussions. The conversation sequences and background information are input into an undesirable thinking monitoring model to obtain the model's output monitoring results. The undesirable thinking monitoring model is trained using sample background information of participants in sample group discussions, sample conversation sequences generated by the sample group discussions, and undesirable thinking labels from the sample group discussions. This overcomes the limitations of traditional methods that cannot comprehensively monitor group discussions and detect undesirable thinking. It enables the monitoring of undesirable thinking in multi-person meetings and provides a reference and basis for resolving undesirable thinking in groups, greatly improving the comprehensiveness and accuracy of decision-making and avoiding the risk of decision failure. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0039] Figure 1 This is a flowchart illustrating the method for monitoring malfunctions based on a large language model provided by the present invention.

[0040] Figure 2 This is a flowchart illustrating the group discussion monitoring and maladaptive thinking detection process provided by the present invention;

[0041] Figure 3 This is a schematic diagram of the structure of the malfunctioning thinking monitoring model provided by the present invention;

[0042] Figure 4 This is a schematic diagram illustrating the training of the malfunction monitoring model provided by this invention;

[0043] Figure 5 This is a flowchart illustrating the anonymous discussion process provided by the present invention;

[0044] Figure 6 This is a flowchart illustrating the nominal group discussion process provided by the present invention;

[0045] Figure 7 This is a schematic diagram of the process for the adequacy check provided by the present invention;

[0046] Figure 8 This is a schematic diagram of the structure of the malthinking monitoring device based on a large language model provided by the present invention;

[0047] Figure 9 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0049] During discussions, due to factors such as participants' lack of ability to propose high-quality solutions, the influence of authority figures, differing interests within and outside the meeting, and time constraints, decisions are often made by the highest decision-maker or by a simple majority rule. This decision-making approach easily leads to undesirable thinking patterns such as dependency, rigidity, and divergent thinking, thus affecting the scientific validity and effectiveness of the decisions. More seriously, this decision-making approach is usually based on partial information, lacking comprehensiveness and depth, and therefore carries a significant risk of decision failure.

[0050] With the continuous development of artificial intelligence technology, Natural Language Processing (NLP) technology has made great strides. However, traditional NLP technology mainly focuses on text summarization, topic extraction, and machine translation, and its ability to handle multi-person, multi-turn dialogues in meetings is relatively limited, making it unable to effectively monitor and identify bad thinking in meetings.

[0051] Furthermore, large-scale models have emerged in recent years, among which large-scale language models have demonstrated natural and powerful dialogue capabilities, providing new directions for machine processing of multi-turn dialogues and offering a feasible path for monitoring undesirable thought processes in groups. However, although some large-scale language models are now capable of processing dialogue content, in multi-person dialogues, especially in multi-person, multi-turn discussions with complex backgrounds, it is necessary to grasp the background of each participant in advance, understand their thought processes and core demands at all times, which makes the monitoring of undesirable thought processes particularly difficult; that is, general-purpose large-scale language models cannot comprehensively control the group discussion process and simultaneously monitor undesirable thought processes.

[0052] In response, this invention provides a method for monitoring maladaptive thinking based on a large-scale language model. The method aims to monitor group discussions using a maladaptive thinking monitoring model built on a large-scale language model, and to detect maladaptive thinking in group discussions based on the information obtained from the monitoring. This enables the discovery of maladaptive thinking in group discussions and provides a reference and basis for the control of maladaptive thinking after it occurs. It can greatly improve the comprehensiveness and accuracy of decision-making, and thus avoid the risk of decision failure.

[0053] Figure 1 This is a flowchart illustrating the method for monitoring undesirable thought processes based on a large language model provided by the present invention. Figure 1 As shown, this method is applied to a conference system, and the method includes:

[0054] Step 110: Monitor the group discussion and obtain background information of the participants and the conversation sequence generated by the group discussion. The conversation sequence includes the participants' speeches, actions and corresponding times.

[0055] Step 120: Input the conversation sequence and background information into the malthought monitoring model to obtain the thought monitoring results output by the malthought monitoring model;

[0056] The negative thought monitoring model is trained based on a large-scale language model, using sample background information of participants in the sample group discussion, sample conversation sequences generated by the sample group discussion, and negative thought labels from the sample group discussion.

[0057] Specifically, considering that traditional natural language processing technology cannot handle multi-person dialogues well and cannot detect undesirable thinking in multi-person meetings, this embodiment of the invention introduces a large-scale language model with strong understanding and interaction capabilities to construct an undesirable thinking monitoring model for multi-person meeting scenarios. This model is used to monitor multi-person meetings / group discussions to detect undesirable thinking, thereby realizing the discovery of undesirable thinking in group discussions and providing a basis for solving undesirable thinking.

[0058] Before that, it is necessary to monitor the entire group discussion process to obtain information such as the speech, identity, and actions of each participant. This will enable monitoring and control of the entire group discussion process, allowing for accurate detection of any undesirable thinking based on the monitored information, and thus obtaining the detection results.

[0059] In other words, once the group discussion has started in the meeting system and all participating parties have entered the discussion, the group discussion can be considered officially started. At this point, the meeting system enters monitoring mode and begins monitoring the entire group discussion. It's important to note that this monitoring goes beyond just tracking the individual participants' speeches; it also includes the time and content of their speeches, their background and identity, and their actions throughout the discussion. This multi-faceted monitoring helps ensure the accuracy of subsequent detection of flawed thinking patterns and the comprehensiveness of group decision-making, thus reducing the risk of decision failure.

[0060] Here, by monitoring group discussions, the conference system can obtain the conversation sequence generated during the group discussions. This conversation sequence includes the speeches and actions of the participants and their corresponding times. That is, by monitoring group discussions, the system can directly obtain the speeches and speaking times of the participants, as well as their actions and their times. However, since this information obtained through direct monitoring is very fragmented, it is difficult for the model to monitor the thinking state of the participants from the fragmented information and detect any undesirable thinking in the group. Therefore, in this embodiment of the invention, the information obtained through direct monitoring needs to be further processed to consolidate the fragmented information. The fragmented information is encoded as special characters and incorporated into the speeches to obtain a multi-person speech record containing complete information, i.e., the conversation sequence of the group discussion.

[0061] It is worth noting that the background information of the participants in the group discussion can be obtained by the meeting system during the group discussion. For example, the background information such as the identity, education, personality traits, and communication skills of the participants can be obtained from their speeches. The background information such as the hands-on ability, past experience, identity information, and hobbies of the participants can also be obtained from their actions. Alternatively, the information can be actively reported / filled out by the participants at the beginning or before the group discussion. Or, the information can be actively collected by the meeting system at the beginning or before the group discussion. This embodiment of the invention does not make specific limitations on this.

[0062] After obtaining the background information of the participants in the group discussion and the conversation sequence generated by the group discussion, in this embodiment of the invention, this information can be pushed to the malthought monitoring model so that the model can detect malthoughts based on the received information, thereby obtaining the detection results of malthoughts in the group discussion, that is, the thought monitoring results.

[0063] It should be noted that during the entire group discussion process, after the group discussion begins, the negative thinking monitoring model will log into the system as an assistant and enter the group discussion to monitor the entire group discussion, detect negative thinking in the group discussion, control the progress of the group discussion, and make comprehensive and accurate decisions.

[0064] Figure 2 This is a flowchart illustrating the group discussion monitoring and maladaptive thinking detection process provided by the present invention, as shown below. Figure 2 As shown, when the meeting system monitors group discussions, it will determine whether the discussion termination conditions are met after each round of discussion. If the conditions are met, the group discussion will be terminated; otherwise, the obtained background information and conversation sequence will be pushed to the negative thinking detection model so that it can detect negative thinking based on the received information and thus obtain the thinking monitoring results.

[0065] Correspondingly, the unhealthy thinking monitoring model continuously receives background information and conversation sequences pushed by the meeting system during group discussions. Based on these two information, it can detect unhealthy thinking in the group discussions, such as dependent thinking, rigid thinking, and divergent thinking. The model can then output the detection results of unhealthy thinking.

[0066] Here, the thought monitoring result is an indication of whether undesirable thoughts have occurred in the group discussion after detection. It can be parameters, commands, warnings, prompts, etc. indicating the occurrence of undesirable thoughts in the group discussion, or it can be operations, prompts, instructions, etc. indicating that undesirable thoughts have not yet occurred in the group discussion. This embodiment of the invention does not specifically limit this.

[0067] Furthermore, it should be noted that when maladaptive thinking is detected during group discussions—that is, when the thinking monitoring results reflect the presence of maladaptive thinking—intervention can be implemented in this embodiment of the invention to ensure the smooth progress of the discussion and to arrive at comprehensive and accurate decisions. This intervention aims to guide the group discussion back to a normal thinking pattern, avoiding the pitfalls of maladaptive thinking that could lead to limited final decisions and a high risk of decision failure. For example, an alert can be generated to remind participants that maladaptive thinking has occurred and to take precautions. Furthermore, control tools can be provided so that participants can access these tools to break free from maladaptive thinking and re-engage to arrive at feasible decisions.

[0068] In this embodiment of the invention, before using the negative thinking monitoring model to monitor negative thinking in group discussions, it is also necessary to pre-train the negative thinking monitoring model by applying the sample background information of the participants in the sample group discussion, the sample conversation sequence generated by the sample group discussion, and the negative thinking labels of the sample group discussion.

[0069] Here, the training process of the maladaptive thinking detection model includes: Considering the powerful understanding and interactive capabilities of large-scale language models, this embodiment of the invention selects to use large-scale language models to process dialogues in multi-person meeting scenarios, leveraging the powerful capabilities of large-scale language models to analyze and understand dialogue content, thereby identifying maladaptive thinking in group discussions. Specifically, a large-scale language model is selected to construct the initial model for the training process. Then, based on pre-collected sample background information of participants in the sample group discussions, sample conversation sequences generated by the sample group discussions, and maladaptive thinking labels from the sample group discussions, the initial model is trained to obtain the trained maladaptive thinking detection model.

[0070] This invention provides a method for monitoring undesirable thinking based on a large-scale language model. It monitors group discussions, acquires background information of participants, and obtains conversation sequences generated during the discussions. These conversation sequences and background information are then input into an undesirable thinking monitoring model to obtain the model's output. The undesirable thinking monitoring model is trained using sample background information of participants in sample group discussions, sample conversation sequences generated by the group discussions, and undesirable thinking labels from the sample group discussions. This overcomes the limitations of traditional methods that cannot comprehensively monitor group discussions and detect undesirable thinking. It enables the monitoring of undesirable thinking in multi-person meetings and provides a reference and basis for resolving undesirable thinking in groups, significantly improving the comprehensiveness and accuracy of decision-making and avoiding the risk of decision failure.

[0071] Based on the above embodiments, the malfunctioning thought monitoring model includes a large language model and a target language model with identical structures, wherein the target language model is constructed based on the large language model; step 120 includes:

[0072] The background information is input into the target language model in the malthinking monitoring model to obtain the background features output by the target language model.

[0073] Background features and conversation sequences are input into a large language model within a maladaptive thinking monitoring model to obtain the thinking monitoring results output by the large language model.

[0074] Considering that large language models often suffer from attention dilution when processing long texts, resulting in poor processing performance, this invention proposes a twin structure when constructing a maladaptive thinking detection model based on a large language model for detecting maladaptive thinking in a group. This symmetrical twin structure can alleviate the attention dilution problem caused by excessively long input text when using a single large language model, thereby improving processing efficiency and performance, and ultimately enhancing the accuracy and precision of maladaptive thinking detection.

[0075] Therefore, in this embodiment of the invention, the structure of the malthought monitoring model for detecting malthought can be set as a twin structure, that is, the malthought monitoring model contains two models with the same structure, namely a large language model and a target language model. To ensure the consistency of their structures, in this embodiment of the invention, the target language model can be constructed based on the large language model.

[0076] In this case, the process of inputting the conversation sequence and background information into the maladaptive thinking detection model so that the model can detect maladaptive thinking and obtain the output of the maladaptive thinking detection model can be achieved by inputting the conversation sequence and background information into two different models in the twin model, so that different models can be used to analyze and process the conversation sequence and background information separately. This not only makes full use of the powerful understanding and analysis capabilities of the large language model, but also alleviates the attention dilution problem when using a single large language model to process long texts, thereby obtaining accurate detection results.

[0077] Specifically, this can involve inputting background information into the target language model of the maladaptive thinking monitoring model, allowing the target language model to analyze and process it to extract the background features corresponding to the background information. Then, these background features and the conversation sequence can be input into the large language model of the maladaptive thinking monitoring model, so that the large language model can perform maladaptive thinking detection on the group discussion based on the two words, in order to detect whether maladaptive thinking such as dependent thinking, rigid thinking, and divergent thinking has occurred in the group discussion, thereby obtaining the thinking monitoring results.

[0078] Based on the above embodiments, the malthought monitoring model further includes a feature fusion layer; the feature fusion layer connects the target language model and the large language model;

[0079] Background features and conversation sequences are input into a large language model within a maladaptive thought monitoring model to obtain the thought monitoring results output by the large language model, including:

[0080] Background features and conversation sequences are input into the feature fusion layer to obtain the fused features output by the feature fusion layer; the fused features are then input into a large language model to obtain the thought monitoring results output by the large language model.

[0081] Specifically, since the large language model and the target language model in the malthought monitoring model are independent of each other, in this embodiment of the invention, a feature fusion layer can be set between the two to connect the large language model and the target language model.

[0082] Figure 3 This is a schematic diagram of the structure of the malfunctioning thinking monitoring model provided by the present invention, as shown below. Figure 3 As shown, in addition to the large language model and the target language model, the bad thinking monitoring model also includes a feature fusion layer that connects the two. This feature fusion layer can be built on the basis of MLP (Multilayer Perceptron).

[0083] The feature fusion layer is used to fuse the output features of the target language model and the conversation sequence, and input the fused features into the input features of the large language model so that the large language model can detect bad thinking in the group discussion and obtain the thinking monitoring results corresponding to the group discussion.

[0084] It's important to note here that large language models contain multiple transformer layers; and since the structure of the large language model is the same as the target language model, the target language model also contains multiple transformer layers. Furthermore, the parameters of the transformer layers in the target language model differ from those in the large language model.

[0085] Correspondingly, the negative thought monitoring model also has multiple feature fusion layers; that is, the number of feature fusion layers is the same as the number of feature transformation layers in the large language model and the target language model. Furthermore, each feature fusion layer connects to the feature transformation layer of the current layer in the target language model and the feature transformation layer of the previous layer in the large language model.

[0086] Specifically, when the background features output by the target language model and the conversation sequence are input into a large language model so that the large language model can detect maladaptive thinking, the background features and conversation sequence can be first input into a feature fusion layer to perform feature fusion and obtain the fused features output by the feature fusion layer. Then, the fused features can be input into the large language model so that the large language model can detect maladaptive thinking in group discussions and obtain the thinking monitoring results output by the large language model.

[0087] Specifically, the background information and conversation sequence are segmented using a tokenizer. The segmented background information is then input into the first feature transformation layer of the target language model to obtain the output features of the first feature transformation layer. Next, this output feature and the segmented conversation sequence are input into the first feature fusion layer to fuse the two and obtain fused features. Then, this fused feature is input into the first feature transformation layer of the large language model to process the fused feature and output the corresponding feature. After that, the output feature and the input features of the second feature transformation layer in the target language model are input into the second feature fusion layer to obtain the second fused feature. This fused feature is then input into the second feature transformation layer of the large language model, and so on, until the output feature of the last feature transformation layer in the large language model is obtained. After passing through a detokenizer, the final thought detection result is obtained.

[0088] In this model, the detokenizer corresponds to the tokenizer model. The tokenizer is used to segment the original text into a series of tokens, while the detokenizer is responsible for recombinating these tokens into meaningful sentences or paragraphs. Specifically, in this embodiment of the invention, the tokenizer is used to segment the conversation sequence and background information, while the detokenizer is used to receive the output of the model and convert it into a specific form.

[0089] In this embodiment of the invention, the feature fusion layer built on top of MLP acts as a mask for the output features of two models with the same structure when performing feature fusion, so that the large language model that processes conversations can make full use of the output features of each feature transformation layer in the target language model; and the hot-swappable architecture can shield the MLP structure and bypass the basic large model at any time, making the model more versatile.

[0090] Based on the above embodiments, the malthinking monitoring model is trained using the following steps:

[0091] Based on a large language model, a twin model is constructed, which includes a first language model and a second language model with the same parameters and structure.

[0092] The sample background information is input into the first language model to obtain the sample background features output by the first language model. The sample background features and sample conversation sequence are input into the second language model to obtain the predictive thinking monitoring results output by the second language model.

[0093] Based on the results of predictive thinking monitoring and negative thinking labels, the first language model in the twin model is trained to obtain the negative thinking monitoring model.

[0094] Specifically, the training process of the negative thought monitoring model includes the following steps:

[0095] To avoid the attention dilution problem that a single large language model might encounter when processing long texts, this embodiment of the invention sets the bad thinking detection model as a twin structure. This allows the two symmetrical models in the twin structure to process background information and conversation sequences respectively, thereby improving processing efficiency and effectiveness. Therefore, during model training, two initial models in the twin structure also need to be pre-built. That is, a twin model can be constructed based on the large language model. This twin model contains two initial models with identical structure and parameters: a first language model and a second language model.

[0096] Specifically, this can be achieved by using a large language model as a benchmark and constructing another large language model with the same structure and parameters. These two models can be referred to as the first language model and the second language model, respectively. For example, the parameters of the large language model can be copied, and a language model can be constructed following its architecture as the first language model. The original large language model can then be used as the second language model. This results in two models with the same structure and parameters, i.e., two initial models. Furthermore, a twin model can be constructed based on these two models.

[0097] Next, the sample background information of the participants in the sample group discussion can be input into the first language model so that the first language model can analyze and process it to obtain the sample background features output by the first language model. This sample background feature, as well as the sample conversation sequence generated by the sample group discussion, can be input into the second language model so that the second language model can monitor negative thinking and obtain the predictive thinking monitoring results output by the second language model.

[0098] Here, the predictive thinking monitoring result is an indication of whether maladaptive thinking has occurred in the sample group discussion after detection. It can be parameters, commands, warnings, prompts, etc. indicating that maladaptive thinking has occurred in the sample group discussion, or it can be operations, prompts, instructions, etc. indicating that maladaptive thinking has not yet occurred in the sample group discussion. This embodiment of the invention does not specifically limit this.

[0099] The sample background information and sample session sequence can be determined in the following ways:

[0100] To achieve malicious thinking detection based on a large language model, enabling the model to detect malicious thinking in group discussions, it is essential to pre-train the model. This training process allows the model to learn the mapping relationship between sample data and labels, enabling it to directly output detection results based on the input information in subsequent applications. Before training the model, training data must be collected, specifically background information and conversation sequences from multiple group discussions. For example, anonymized data from multiple group discussions can be obtained from the conference system's database. This anonymized data for each group discussion includes the background of each participant, their statements, their various operations related to the conference or system during the discussion, and their speaking and operation times.

[0101] Furthermore, after obtaining anonymized data from multiple group discussions, this data needs to be processed to form training sample data and labels. Specifically, the anonymized data can be processed to convert it into text format and obtain the speech of each participant. For example, for audio files from group discussions, ASR (Automatic Speech Recognition) technology can be used to convert them into text. Based on this, the participants' actions and speaker annotation techniques can be used to separate and annotate speakers, obtaining the speech text of each participant. Then, the data can be cleaned, with operations and their corresponding times used as special characters and incorporated into the speech context to obtain the participants' speech records, i.e., the sample conversation sequence. Correspondingly, based on the anonymized data, a participant ID (IdentityDocument)-background dictionary, i.e., the sample background information, can be obtained.

[0102] After obtaining the predictive thinking monitoring results output by the twin model, the model can be trained based on these results to obtain a trained model for monitoring maladaptive thinking.

[0103] Figure 4 This is a schematic diagram of the training of the malfunction monitoring model provided by the present invention, such as... Figure 4 As shown, based on the predictive thinking monitoring results output by the second language model and the negative thinking labels, the parameters of the first language model in the twin model can be iterated so that the features output by the first language model after parameter adjustment, together with the features output by the second language model, can be used to detect negative thinking as consistent as possible with the negative thinking labels. Finally, a trained negative thinking monitoring model can be obtained.

[0104] It should be noted that the training of the twin model adopts the LoRA fine-tuning method. During the fine-tuning process, the parameters of the second language model are frozen and do not participate in the training. The first language model is fine-tuned using LoRA technology. During fine-tuning, the sample background information of a specific person and the corresponding speech record are used as input. The speech record of the next person in the sample conversation sequence of that specific person (including the Assistant's speech, i.e., containing the negative thought label) is used as the supervision signal to fine-tune the first language model. Finally, the trained first language model can be obtained. On this basis, the negative thought monitoring model can be constructed.

[0105] The labels for negative thinking patterns can be manually annotated by thinking experts after obtaining the sample conversation sequence, or automatically annotated using a general large model. Specifically, during the annotation process, assistant comments can be inserted at points in the sample conversation sequence where negative thinking patterns such as forced substitution, disconnect between different levels, language communication barriers, groupthink, blindly applying experience, failing to distinguish between assumptions and facts, or excessively sarcastic thinking occur, indicating the presence of negative thinking patterns at those points, thereby obtaining the corresponding negative thinking pattern labels for the sample group discussion.

[0106] Based on the above embodiments, the first language model includes multiple first feature conversion layers, and the second language model includes multiple second feature conversion layers;

[0107] The twin model also includes multiple initial feature fusion layers. The first language model and the second language model are connected through multiple initial feature fusion layers. The first feature transformation layer, the second feature transformation layer and the initial feature fusion layer have the same number of layers.

[0108] Any initial feature fusion layer is connected to the first feature transformation layer of the corresponding layer in the first language model and the second feature transformation layer of the layer above the corresponding layer in the second language model;

[0109] The initial feature fusion layer is used to fuse the output features of the first feature transformation layer of the corresponding layer in the first language model with the output features of the second feature transformation layer of the previous layer in the second language model, and use the fused features as the input features of the second feature transformation layer of the corresponding layer in the second language model.

[0110] Specifically, see Figure 4 As can be seen, in this embodiment of the invention, the twin model also includes multiple initial feature fusion layers, which are constructed on the basis of MLP. The initial feature fusion layers are used to connect the first language model and the second language model in the twin model. The first language model contains multiple first feature transformation layers, and the second language model contains multiple second feature transformation layers; furthermore, the number of layers in the twin model is the same for the first feature transformation layers, the second feature transformation layers, and the initial feature fusion layer.

[0111] In this model, any initial feature fusion layer, when connecting the first language model and the second language model, connects the first feature transformation layer of the corresponding layer in the first language model and the second feature transformation layer of the layer above the corresponding layer in the second language model. In short, the second / second initial feature fusion layer connects the second / second first feature transformation layer in the first language model of the twin model to the first / first second feature transformation layer in the second language model; the third / third initial feature fusion layer connects the third / third first feature transformation layer in the first language model of the twin model to the second / second second feature transformation layer in the second language model. The first / first initial feature fusion layer connects the first / first first feature transformation layer in the first language model of the twin model to the tokenizer of the sample conversation sequence.

[0112] Furthermore, each initial feature fusion layer fuses the output features of the first feature transformation layer of the corresponding layer in the first language model and the output features of the second feature transformation layer of the layer above the corresponding layer in the second language model to obtain the fused features of the corresponding layer. These fused features are then used as the input features of the second feature transformation layer of the corresponding layer in the second language model. In other words, the initial feature fusion layer of the current layer is used to fuse the output features of the first feature transformation layer of the current layer and the output features of the second feature transformation layer of the layer above.

[0113] Based on the above embodiments, the first language model in the twin model is trained based on the predictive thinking monitoring results and negative thinking labels to obtain the negative thinking monitoring model, including:

[0114] Based on the results of predictive thinking monitoring and negative thinking labels, the first language model and each initial feature fusion layer in the twin model are trained to obtain the target language model and multiple feature fusion layers.

[0115] A model for monitoring negative thinking is constructed based on a large-scale language model, a target language model, and various feature fusion layers.

[0116] Specifically, the process of training the first language model in the twin model based on the predictive thinking monitoring results and negative thinking labels to obtain the negative thinking monitoring model can include:

[0117] First, the twin model is trained using the predictive thinking monitoring results and negative thinking labels output by the twin model. Specifically, the parameters of the second language model are frozen during training, while the first language model and the initial feature fusion layer are fine-tuned using LoRA technology. That is, based on the predictive thinking detection results and negative thinking labels, the first language model and each initial feature fusion layer in the twin model are trained to adjust their parameters. This ensures that the predicted thinking monitoring results output by the twin model for the input sample group discussion sample background information and sample conversation sequence are as close as possible to the negative thinking labels of the sample group discussion, thus obtaining the trained model and fusion layers, namely the target language model and each feature fusion layer.

[0118] Subsequently, based on this target language model and the various feature fusion layers, a malthought detection model can be constructed. Specifically, this can be achieved by combining the target language model and the various feature fusion layers with a large language model to obtain the final malthought detection model. Here, since the structure of the initial model corresponding to the target language model is the same as the structure of the first language model, and the first language model is the large language model, the final malthought detection model contains a symmetrical large language model and target language model, as well as multiple feature fusion layers connecting the two.

[0119] Furthermore, after obtaining the malthought monitoring model, this embodiment of the invention can also perform model evaluation to verify the performance of the trained malthought monitoring model. This embodiment proposes two evaluation methods: First, the model predicts the next person's speech record based on the current person's speech record, and the prediction result and the test set ground truth are input into a large language model. The similarity is calculated by taking the output of the last n / 2 hidden layers of the large language model. Second, the model predicts the next person's speech record based on a portion of the preceding speech record, and the prediction result is added to the historical record, iterating continuously until the end. The conversation sequence generated by the model, the conversation sequence labeled in the test set, and the background information are input into the large model to evaluate the model's performance.

[0120] Based on the above embodiments, the conversation sequence and background information are input into the malfunctioning thought monitoring model to obtain the thought monitoring results output by the malfunctioning thought monitoring model, and then the process further includes:

[0121] When the results of the thought monitoring indicate the presence of negative thoughts in the group discussion, an alarm prompt and control tool for negative thoughts are obtained so that the participants in the group discussion can control the group discussion based on the alarm prompt and control tool; the alarm prompt and control tool for negative thoughts are generated by the negative thought monitoring model based on the thought monitoring results.

[0122] Specifically, in step 120, after inputting the conversation sequence and background information into the maladaptive thinking monitoring model and obtaining the thinking monitoring results output by the model, if maladaptive thinking is detected in the group discussion, that is, the thinking monitoring results indicate that maladaptive thinking has occurred in the group discussion, then in order to ensure that the discussion can reach a comprehensive and accurate decision, the group discussion can be intervened. This can be done by obtaining maladaptive thinking warning prompts and control tools, so that the participants in the group discussion can know that maladaptive thinking has occurred in the discussion through the maladaptive thinking warning prompts, and can make timely adjustments. Furthermore, the group discussion can be controlled through the control tools, thereby getting rid of the predicament of maladaptive thinking, returning to a normal thinking mode, and re-discussing to obtain a comprehensive, accurate, and reliable decision.

[0123] Among them, the negative thinking warning and control tools can be generated by the negative thinking monitoring model when it detects negative thinking in the discussion, and can be output along with the thinking monitoring results reflecting the occurrence of negative thinking in the group discussion.

[0124] In detail, in the embodiments of the present invention, the control tools provided by the model can be of various types, such as anonymous seminars, nominal groups, group discussions, seminar adequacy checks, etc.

[0125] The anonymous discussion tool is provided by the meeting system to avoid potential influence from individuals outside the discussion. Figure 5 This is a flowchart illustrating the anonymous discussion process provided by the present invention, such as... Figure 5 As shown, when participants initiate anonymous discussions or the model detects erroneous thinking, the meeting system will send anonymous invitations on behalf of the system. Participants can choose to enter anonymous mode. When all participants are anonymous, the non-anonymous status will no longer be displayed; otherwise, those who have entered anonymous mode will appear as incremental participants to ensure the effectiveness of anonymity. Participants are indexed by a unique identifier and displayed as random IDs in group discussions. For audio and video conferencing, virtual digital human technology replaces the original participants' appearance and voice.

[0126] The nominal group tool is designed to avoid misinterpreting the views of those who remain silent. The meeting system provides a nominal group function. Figure 6 This is a flowchart illustrating the nominal group discussion process provided by the present invention, such as... Figure 6 As shown, when participants initiate a nominal group discussion or the model detects maladaptive thinking, the meeting system will issue a nominal group invitation on behalf of the system. Each participant will be asked to express their views and reasons on the discussion topic. To avoid individual opinions taking up too much time, the nominal group mode will limit the speaking time limit for each participant.

[0127] The group discussion tool is designed to alleviate the pressure of individual opposition to group opinions. The meeting system provides group discussion functionality, which inherits multiple features from the entire meeting system, including anonymous discussion. Participants can end the discussion at any time and return to the original group discussion. When any participant in a group discussion feels pressured by potential opposition, they can initiate a group discussion as a system user. Participants enter the group anonymously by default, while remaining online in the original group discussion, ensuring that the pressure from the original group discussion does not leak to the current group.

[0128] The tool for checking the adequacy of discussions is as follows: To avoid reaching a consensus too quickly without in-depth discussion of the issues, a discussion adequacy check can be performed. Figure 7 This is a schematic diagram of the adequacy check process provided by the present invention, such as... Figure 7 As shown, the topic tool first checks the relevance of participants' speeches to the topic and the completeness of their arguments. Irrelevant arguments are removed from the argument set. When the completeness of the argument set is sufficient to support the root argument, the current discussion is considered sufficient, and the discussion can be ended or the topic changed; otherwise, the topic tool will obtain permission to prevent the discussion from ending or continuing.

[0129] In this embodiment of the invention, when the thinking monitoring results indicate the presence of undesirable thinking in a group discussion, an alarm prompt and control tool for undesirable thinking are obtained to regulate the group discussion. For the first time, the regulation of undesirable thinking by experts is applied to online group discussions / conferences in the form of an assistant through a large language model, providing a specific and feasible path for solving undesirable thinking in online group discussions / conferences.

[0130] The following describes the malthought monitoring device based on a large language model provided by the present invention. The malthought monitoring device based on a large language model described below can be referred to in correspondence with the malthought monitoring method based on a large language model described above.

[0131] Figure 8 This is a schematic diagram of the structure of the malthought monitoring device based on a large language model provided by the present invention, as shown below. Figure 8 As shown, this device is used in a conference system and includes:

[0132] The acquisition unit 810 is used to monitor the group discussion and acquire the background information of the participants in the group discussion, as well as the conversation sequence generated by the group discussion. The conversation sequence includes the speeches and operations of the participants in the group discussion and their corresponding times.

[0133] The monitoring unit 820 is used to input the conversation sequence and the background information into the negative thinking monitoring model to obtain the thinking monitoring results output by the negative thinking monitoring model.

[0134] The negative thought monitoring model is trained based on a large language model, using the sample background information of participants in the sample group discussion, the sample conversation sequence generated by the sample group discussion, and the negative thought labels of the sample group discussion.

[0135] This invention provides a device for monitoring undesirable thinking based on a large-scale language model. It monitors group discussions, acquires background information of participants, and obtains conversation sequences generated during the discussions. The conversation sequences and background information are input into the undesirable thinking monitoring model to obtain the monitoring results output by the model. The undesirable thinking monitoring model is trained using sample background information of participants in sample group discussions, sample conversation sequences generated by the sample group discussions, and undesirable thinking labels from the sample group discussions. This overcomes the shortcomings of traditional solutions that cannot comprehensively monitor group discussions and detect undesirable thinking. It enables the monitoring of undesirable thinking in multi-person meetings and provides a reference and basis for resolving undesirable thinking in groups, greatly improving the comprehensiveness and accuracy of decision-making and avoiding the risk of decision failure.

[0136] Based on the above embodiments, the malfunctioning thought monitoring model includes a large language model and a target language model with identical structures, wherein the target language model is constructed based on the large language model; the monitoring unit 820 is used for:

[0137] The background information is input into the target language model in the negative thought monitoring model to obtain the background features output by the target language model.

[0138] The background features and the conversation sequence are input into the large language model in the malthought monitoring model to obtain the thought monitoring results output by the large language model.

[0139] Based on the above embodiments, the malthought monitoring model further includes a feature fusion layer; the feature fusion layer connects the target language model and the large language model;

[0140] Monitoring unit 820 is used for:

[0141] The background features and the conversation sequence are input into the feature fusion layer to obtain the fused features output by the feature fusion layer;

[0142] The fusion features are input into the large language model to obtain the thought monitoring results output by the large language model.

[0143] Based on the above embodiments, the device further includes a training unit for:

[0144] Based on the large language model, a twin model is constructed, which includes a first language model and a second language model with the same parameters and structure.

[0145] The sample background information is input into the first language model to obtain the sample background features output by the first language model. The sample background features and the sample conversation sequence are input into the second language model to obtain the predictive thinking monitoring results output by the second language model.

[0146] Based on the predicted thinking monitoring results and the negative thinking labels, the first language model in the twin model is trained to obtain the negative thinking monitoring model.

[0147] Based on the above embodiments, the first language model includes multiple first feature conversion layers, and the second language model includes multiple second feature conversion layers;

[0148] The twin model also includes multiple initial feature fusion layers. The first language model and the second language model are connected through multiple initial feature fusion layers. The first feature transformation layer, the second feature transformation layer and the initial feature fusion layer have the same number of layers.

[0149] Any initial feature fusion layer connects the first feature transformation layer of the corresponding layer in the first language model to the second feature transformation layer of the layer above the corresponding layer in the second language model;

[0150] The initial feature fusion layer is used to fuse the output features of the first feature transformation layer of the corresponding layer in the first language model with the output features of the second feature transformation layer of the previous layer in the second language model, and use the fused features as the input features of the second feature transformation layer of the corresponding layer in the second language model.

[0151] Based on the above embodiments, the training unit is used for:

[0152] Based on the predicted thinking monitoring results and the negative thinking labels, the first language model and each initial feature fusion layer in the twin model are trained to obtain the target language model and multiple feature fusion layers.

[0153] Based on the large-scale language model, the target language model, and the various feature fusion layers, a model for monitoring negative thinking is constructed.

[0154] Based on the above embodiments, the device further includes a control unit, used for:

[0155] When the thought monitoring results indicate the presence of negative thoughts in the group discussion, an alarm prompt and control tool for negative thoughts are obtained so that the participants in the group discussion can control the group discussion based on the alarm prompt and control tool for negative thoughts;

[0156] The negative thought warning and control tool is generated by the negative thought monitoring model based on the thought monitoring results.

[0157] Figure 9 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 9 As shown, the electronic device may include: a processor 910, a communication interface 920, a memory 930, and a communication bus 940, wherein the processor 910, the communication interface 920, and the memory 930 communicate with each other through the communication bus 940. The processor 910 can call logical instructions in the memory 930 to execute a method for monitoring unhealthy thinking based on a large language model. This method includes: monitoring group discussions and obtaining background information of the participants in the group discussions, as well as the conversation sequences generated by the group discussions, the conversation sequences containing the participants' speeches, operations, and corresponding times; inputting the conversation sequences and the background information into an unhealthy thinking monitoring model to obtain the thinking monitoring results output by the unhealthy thinking monitoring model; the unhealthy thinking monitoring model is trained based on a large language model, using sample background information of the participants in the sample group discussions, sample conversation sequences generated by the sample group discussions, and unhealthy thinking labels of the sample group discussions.

[0158] Furthermore, the logical instructions in the aforementioned memory 930 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0159] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, wherein when the program instructions are executed by a computer, the computer is able to execute the method for monitoring maladaptive thinking based on a large language model provided by the above methods, the method comprising: monitoring a group discussion and obtaining background information of the participants in the group discussion and a conversation sequence generated by the group discussion, the conversation sequence including the participants' speeches, operations and their corresponding times; inputting the conversation sequence and the background information into a maladaptive thinking monitoring model to obtain a thinking monitoring result output by the maladaptive thinking monitoring model; the maladaptive thinking monitoring model is trained on the basis of a large language model by applying sample background information of the participants in the sample group discussion, sample conversation sequences generated by the sample group discussion, and maladaptive thinking labels of the sample group discussion.

[0160] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method for monitoring maladaptive thinking based on a large language model provided by the above methods. This method includes: monitoring a group discussion and obtaining background information of the participants in the group discussion, as well as a conversation sequence generated by the group discussion, the conversation sequence containing the participants' speeches, actions, and corresponding times; inputting the conversation sequence and the background information into a maladaptive thinking monitoring model to obtain a thinking monitoring result output by the maladaptive thinking monitoring model; the maladaptive thinking monitoring model is trained based on a large language model, using sample background information of the participants in the sample group discussion, sample conversation sequences generated by the sample group discussion, and maladaptive thinking labels from the sample group discussion.

[0161] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0162] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0163] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for monitoring maladaptive thinking based on a large-scale language model, characterized in that, Applied to a conference system, the method includes: The group discussion is monitored, and background information of the participants in the group discussion and the conversation sequence generated by the group discussion are obtained. The conversation sequence includes the speeches, actions and corresponding times of the participants in the group discussion. The conversation sequence and the background information are input into the negative thought monitoring model to obtain the thought monitoring results output by the negative thought monitoring model. The negative thought monitoring model is trained based on a large language model, using the sample background information of the participants in the sample group discussion, the sample conversation sequence generated by the sample group discussion, and the negative thought labels of the sample group discussion. The negative thought monitoring model includes a large language model and a target language model with identical structures, and the target language model is constructed based on the large language model. The step of inputting the conversation sequence and the background information into the malfunctioning thought monitoring model to obtain the thought monitoring results output by the malfunctioning thought monitoring model includes: The background information is input into the target language model in the negative thought monitoring model to obtain the background features output by the target language model. The background features and the conversation sequence are input into the large language model in the malthought monitoring model to obtain the thought monitoring results output by the large language model; The negative thought monitoring model is trained based on the following steps: Based on the large language model, a twin model is constructed, which includes a first language model and a second language model with the same parameters and structure. The sample background information is input into the first language model to obtain the sample background features output by the first language model. The sample background features and the sample conversation sequence are input into the second language model to obtain the predictive thinking monitoring results output by the second language model. Based on the predicted thinking monitoring results and the negative thinking labels, the first language model in the twin model is trained to obtain the negative thinking monitoring model. The first language model includes multiple first feature transformation layers, and the second language model includes multiple second feature transformation layers; The twin model also includes multiple initial feature fusion layers. The first language model and the second language model are connected through multiple initial feature fusion layers. The first feature transformation layer, the second feature transformation layer and the initial feature fusion layer have the same number of layers. Any initial feature fusion layer connects the first feature transformation layer of the corresponding layer in the first language model to the second feature transformation layer of the layer above the corresponding layer in the second language model; The initial feature fusion layer is used to fuse the output features of the first feature transformation layer of the corresponding layer in the first language model with the output features of the second feature transformation layer of the previous layer in the second language model, and use the fused features as the input features of the second feature transformation layer of the corresponding layer in the second language model.

2. The method for monitoring maladaptive thinking based on a large-scale language model according to claim 1, characterized in that, The negative thought monitoring model also includes a feature fusion layer; The feature fusion layer connects the target language model and the large language model; The process of inputting the background features and the conversation sequence into the large language model of the maladaptive thinking monitoring model to obtain the thinking monitoring results output by the large language model includes: The background features and the conversation sequence are input into the feature fusion layer to obtain the fused features output by the feature fusion layer; The fusion features are input into the large language model to obtain the thought monitoring results output by the large language model.

3. The method for monitoring maladaptive thinking based on a large-scale language model according to claim 1, characterized in that, The step of training the first language model in the twin model based on the predicted thinking monitoring results and the negative thinking labels to obtain the negative thinking monitoring model includes: Based on the predicted thinking monitoring results and the negative thinking labels, the first language model and each initial feature fusion layer in the twin model are trained to obtain the target language model and multiple feature fusion layers. Based on the large-scale language model, the target language model, and the various feature fusion layers, a model for monitoring negative thinking is constructed.

4. The method for monitoring maladaptive thinking based on a large-scale language model according to claim 1 or 2, characterized in that, The step of inputting the conversation sequence and the background information into the malfunction thinking detection model to obtain the thinking monitoring results output by the malfunction thinking detection model further includes: When the thought monitoring results indicate the presence of negative thoughts in the group discussion, an alarm prompt and control tool for negative thoughts are obtained so that the participants in the group discussion can control the group discussion based on the alarm prompt and control tool for negative thoughts; The negative thought warning and control tool is generated by the negative thought monitoring model based on the thought monitoring results.

5. A device for monitoring unhealthy thinking based on a large-scale language model, characterized in that, The device, used in a conference system, includes: The acquisition unit is used to monitor the group discussion and acquire the background information of the participants in the group discussion, as well as the conversation sequence generated by the group discussion. The conversation sequence includes the speeches and operations of the participants in the group discussion and their corresponding times. The monitoring unit is used to input the background information into the target language model in the malthought monitoring model to obtain the background features output by the target language model; input the background features and the conversation sequence into the large language model in the malthought monitoring model to obtain the thought monitoring results output by the large language model; the large language model and the target language model have the same structure, and the target language model is constructed based on the large language model; The negative thought monitoring model is trained based on a large language model, using the sample background information of the participants in the sample group discussion, the sample conversation sequence generated by the sample group discussion, and the negative thought labels of the sample group discussion. The negative thought monitoring model is trained based on the following steps: Based on the large language model, a twin model is constructed. The twin model includes multiple initial feature fusion layers, as well as a first language model and a second language model with the same parameters and structure. The first language model and the second language model are connected by multiple initial feature fusion layers. The first feature transformation layer in the first language model, the second feature transformation layer in the second language model, and the initial feature fusion layer have the same number of layers. The sample background information is input into the first language model to obtain the sample background features output by the first language model. The sample background features and the sample conversation sequence are input into the second language model to obtain the predictive thinking monitoring results output by the second language model. Based on the predicted thinking monitoring results and the negative thinking labels, the first language model in the twin model is trained to obtain the negative thinking monitoring model. Any initial feature fusion layer connects the first feature transformation layer of the corresponding layer in the first language model to the second feature transformation layer of the layer above the corresponding layer in the second language model; The initial feature fusion layer is used to fuse the output features of the first feature transformation layer of the corresponding layer in the first language model with the output features of the second feature transformation layer of the previous layer in the second language model, and use the fused features as the input features of the second feature transformation layer of the corresponding layer in the second language model.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method for monitoring malthought based on a large language model as described in any one of claims 1 to 4.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for monitoring malthought based on a large language model as described in any one of claims 1 to 4.

Images